Does Machine Bring in Extra Bias in Learning? Approximating Fairness in Models Promptly

• DOI: 10.48550/arxiv.2405.09251
• Publication Date: 2024-05-15
• Authors: Yijun Bian; Yujie Luo
• Keywords: I.2; FOS: Computer and information sciences; Computer Science - Machine Learning; Computer Science - Computers and Society; I.2.6; Computers and Society (cs.CY); K.4.2; I.2.0; 68T01, 68T09, 68T20; Machine Learning (cs.LG); I.2; I.2.6; I.2.0; K.4.2

Providing various machine learning (ML) applications in the real world, concerns about discrimination hidden ML models are growing, particularly high-stakes domains. Existing techniques for assessing level of include commonly used group and individual fairness measures. However, these two types measures usually hard to be compatible with each other, even different might incompatible as well. To address this issue, we investigate evaluate classifiers from a manifold perspective propose "harmonic measure via manifolds (HFM)" based on distances between sets. Yet direct calculation too expensive afford, reducing its practical applicability. Therefore, devise an approximation algorithm named "Approximation distance sets (ApproxDist)" facilitate accurate estimation distances, further demonstrate algorithmic effectiveness under certain reasonable assumptions. Empirical results indicate that proposed HFM is valid ApproxDist effective efficient.